Shubham Gupta

Machine Learning, Computer Vision, Robotics

Download Resume

About Me

Hi, I am currently pursuing MS in Robotics at University of Pennsylvania. I did my undergraduation in Mechatronics Engineering from Thapar University, India. Since beginning my journey in Robotics, I have been involved in the development of various projects. My interests lie in Machine Vision and Intelligence. Take a moment to check my portfolio, and get in touch with any questions.


Robotics Research Center, IIIT Hyderabad

Research Assistant

I got interested in Perception and State Estimation during my time at IIIT Hyderabad. I studied various Visual Inertial algorithms such as ROVIO, VINS-MONO, and MSCKF-VIO and implemented the ROS packages on monocular and stereo cameras. I also experimented with an approach to estimate the trajectory of a camera by identifying the static objects in a scene. I also implemented the full structure from motion pipeline to compute camera poses and 3D point cloud simultaneously.

Programme for Autonomous Robotics, IIT Delhi

Intern, Video

I spent developing a mobile robot capable of being operated remotely or moving autonomously in a pre-defined path. The robot could detect obstacles, recognise human faces and interact with humans with the help of touch screen and speaker. The work involved CAD modelling, hardware and firmware design, odometry and face recognition implementation. Please find a video here


University of Pennsylvania

January 2019 - December 2021

MSE Robotics Engineering, Current GPA: 3.95/4

Relevant coursework: Applied Machine Leaning, Learning in Robotics, Machine Perception, Computer Vision and Computational Photography, Advances Topic in Machine Perception, Linear Systems, Advanced Robotics

Thapar University

August 2014 - May 2018

Bachelor of Engineering in Mechatronics Engineering, GPA: 8.35/10

Relevant coursework: Control Systems, Advanced Control Systems, Robotics Engineering, Signals and Systems, Digital Signal Processing, Kinematics and Dynamics of Machines


Instance Segmentation

  • Implemented a single-stage instance segmentation pipeline "Segmenting Objects by Location" as part of course CIS 680: Advanced Machine Perception, MAP value achieved: 0.46
  • Dataset: Subset of COCO dataset with 10,000 images comprising of People, Vehicles, and Animals
  • Backbone used: RESNET 50 FPN

Object Detection: Faster RCNN

  • MAP value achieved: 0.76
  • Dataset: Subset of COCO dataset with 10,000 images comprising of People, Vehicles, and Animals
  • Trained a lighter version of backbone and Region Proposal Network for the first stage and box regresor and classifier for the second stage
  • Backbone used for second stage: RESNET 50 FPN

Object Detection: YOLO

  • MAP Value achieved: 0.46
  • Dataset: 10,000 street scene images with pedestrians, cars, and traffic lights labels
  • Implemented the whole pipeline from scratch and compared the performance of classifier in case of different targets for confidence loss

Particle Filter based SLAM

  • Simultaneous Localization and Mapping from LIDAR scan and IMU data
  • A local map is built for every particle by extending LIDAR scan from particle's position(estimated from odometry information after adding random noise)
  • Update of each particle's weight is done based on correlation of local map with best estimate of map

Optical Flow

  • Tracking pre-initialized bounding boxes in a video
  • KLT tracking algorithm to calculate the Optical Flow
  • Similarity Transformation estimation b/w the tracked points in two frames for calculating bounding box movement